Instrumental methods in detecting taints

3 Instrumental methods in detecting taints

  and off-flavours

W. J. Reid, Leatherhead Food International, UK

3.1 Introduction

  The instrumental analysis of taints and off-flavours is a complementary technique to that of sensory analysis and presents its own interesting blend of certainties and challenges, largely governed by the sensory characteris- tics of the compounds under study. The presence of a tainting compound in

  a sample is apparent from the change in odour or flavour. The analyst begins, therefore, by knowing there is something to find. The description of the taint provides an additional parameter, because any target compound identified in the sample must have the same taste and odour characteristics as those derived from sensory analysis. Many of the compounds that cause taint do so because they are very potent and can be perceived at extremely low concentrations by human senses. For example, 2,6-dibromophenol has

  a taste threshold in water of 5 ¥ 10 -1 mg l (Whitfield et al., 1988). This pre- sents a challenge to the analyst, given the need to detect these very low con-

  centrations of the tainting compound, often in the presence of much higher concentrations of other naturally occurring materials from the sample. This threshold data is also a criterion for the analysis, since any candidate com- pound identified by the chemical analysis must be present in the sample at

  a high enough concentration to cause the perceived taint.

  Generally, taint analysis falls into one of two categories. In one case, the compounds causing the taint are known and the analysis can be optimised for those compounds and the sample matrix. In the other case, the nature of the tainting compounds is not known and a more general approach is needed, often using sensory data in tandem with the instrumental analysis.

  32 Taints and off-flavours in food used for tainting compounds in food. It will start with solvent-based

  extraction techniques, for example, liquid–liquid extraction, steam distilla- tion or combined steam distillation and solvent extraction. The next section will deal with headspace methods such as static and dynamic headspace sampling and closed loop stripping. This will be followed by a description of the relatively recent technique of solid phase microextraction (SPME) and some applications of the technique to taint analysis. The next section will consider instrumental analysis, including combined gas chromatogra- phy–olfactometry (GCO), gas chromatography with selective detectors and gas chromatography–mass spectrometry (GCMS), while the final section will cover new methods in taint analysis such electronic noses and stir-bar sorbtive extraction (SBSE).

3.2 Liquid-based extraction techniques

  The methods used for the extraction and concentration of tainting com- pounds are similar to those used for the extraction of other organic com- pounds and, in particular, those used for flavour compounds (Wilkes et al., 2000). The difference lies in the need to detect the very low concentrations of the target compounds that can cause taint in the sample. The aim of the extraction is to arrive at a solution of the tainting compounds in a form that can be introduced into the analytical instrument. It is often useful to work back from the requirements of the instrument in order to define the requirements of the extraction.

  Consider the situation where a specific compound is to be analysed, using

  a detection system such as high resolution gas chromatography (HRGC) coupled to a sensitive detector such as a mass spectrometer (MS) working in the selected ion monitoring mode (SIM) or, for halogenated compounds, an electron capture detector (ECD). It is reasonable to suppose that a peak with a good signal-to-noise ratio could be obtained from 1 pg of material injected into the instrument. If the final volume of the solution obtained from the extraction was 0.2 ml and the injection volume was 1 ml, then the extract will need to contain around 200 pg of material. It follows, therefore,

  that to detect 2,4,6-trichloroanisole at its taste threshold of 0.01 mg l -1 in wine (Buser et al., 1982) it would be necessary to extract 20 ml of wine. If we aim

  for a detection limit at a concentration of one hundredth of the taste thre- shold, then it would be necessary to extract 2000 ml of wine. On the other hand, setting a detection limit for 2-bromophenol at one hundredth of its

  taste threshold in prawns, which is 2 mg kg -1 (Whitfield et al., 1988), would require the extraction of 10 g of prawns. In cases where the aim is to iden-

  tify an unknown compound it may be necessary to extract greater amounts of sample. It is apparent that any extraction procedure requires the removal of material that might interfere with the analysis, particularly given the extreme concentration factor required.

3.2.1 Solvent extraction

  The simplest method of extracting organic compounds from any matrix is solvent extraction. In some cases the extract can be used directly for analy- sis. Indole and skatole have been associated with a taint in the flesh of intact male pigs. A high-performance liquid chromatographic (HPLC) method for these compounds in pig fat and meat has been developed in which a simple solvent extract is applied directly to the HPLC system (Garcia-Reguero and Ruis, 1998; Ruis and Garcia-Reguero, 2001), using a fluorescence detector which is selective enough not to respond to any co-eluting material. More usually, however, the extraction stage is followed by a series of clean-up stages to prepare the extract for analysis, often using the sensory charac- teristics of the tainting compounds to guide the procedure. Whitfield and co-workers (Whitfield et al., 1991) used Soxhlet extraction with pentane to examine an earthy taint in flour. They extracted 250 g of flour, concentrated the pentane extract and subjected it to simultaneous steam distillation and extraction (SDE) to recover the volatile compounds. The GCMS analysis was used to identify the tainting compound as geosmin.

  An investigation into a medicinal taint in cake mix (Sevenants and Sanders, 1984) began with a solvent extract of the material. The extract was then partitioned between different solvents and the fraction containing the medicinal taint was separated into basic, neutral and acidic fractions. The acidic portion, which had the medicinal smell, was analysed by HRGC using an olfactometer and a flame ionisation detector (FID) in parallel. The extract was re-analysed by GCMS in an attempt to obtain a spectrum at the retention time identified from the olfactometry measurements. In this case the GCMS data proved to be complex around the region of interest.The analysis was simplified by the discovery that a mixture of minor ingredients of the cake mix produced the same taint. Extraction, clean-up and analysis of this mixture resulted in the identification of 2-iodo-4- methylphenol, which was shown to be formed by the reaction between cresol in flavouring and iodine from sea salt, two of the components of the mixture. A similar procedure was used to identify 2,6-dibromophenol in prawns (Whitfield et al., 1988). The prawns were macerated in water and the mixture extracted in a liquid–liquid extractor in order to produce enough material for the identification of the compound. Subsequent target analyses used less material and relied on SDE for the sample preparation.

3.2.2 Supercritical fluid extraction

  Under elevated pressures and temperatures, compounds that are normally gases reach a supercritical state in which they possess properties charac- teristic of both gases and liquids. In particular they have the penetrative power of a gas and the solvating power of a liquid. Supercritical fluid extrac- tion (SFE) is widely used in food analysis and has been extensively reviewed (Anklam et al., 1998; Chester et al., 1998; Camel, 1998; Sihvonen

  Instrumental methods in detecting taints and off-flavours 33

  34 Taints and off-flavours in food et al., 1999). In a recent book, Mukhopadhyay (2000) provides a detailed

  discussion on the theoretical aspects of SFE.

  The fat of intact male pigs can have an objectionable odour. Androstenone and skatole have been reported to contribute to this taint. In an effort to eliminate the use of the solvents required for a normal liquid extraction, SFE was studied as an extraction method for these compounds (Zablotsky et al., 1993; Zablotsky et al., 1995). The authors used a static extraction system and found that they could achieve a recovery of around

  97 for androstenone and around 65 for skatole. Although the recovery of skatole was low, it was found to be reproducible. A dynamic extraction system was also used to extract androstenone from pig fat (Magard et al., 1995). Here the supercritical fluid passed through the sample and was led out from the apparatus through a trap containing octadecyl silica. After extraction the trap was eluted with 1.5 ml of dichloromethane to recover the compound. The various parameters governing the extraction were studied in order to maximise the recovery of androstenone and minimise the extraction of fat. It was found that the solubility of the fat increased with temperature and pressure and so the lowest possible temperature and pressure, consistent with good recovery of the compound, were used. The analysis using SFE was compared with those from two standard methods based on radioimmunoassay and HPLC and the results from the three methods were found to be comparable.

  A perennial problem for wine producers is the ‘cork taint’ in wine, which is reported to be the result of contamination by 2,4,6-trichloroanisole (Buser et al., 1982) which originates in cork stoppers. A dynamic SFE-based method was developed for the analysis of 2,4,6-trichloroanisole in cork (Taylor et al., 2000) with the aim of eliminating the solvent used in Soxhlet extraction and reducing the time of the analysis. After extraction the com- pound was eluted from the trap in methanol and analysed by GCMS. The results from SFE were compared with those from Soxhlet extraction using methanol and found to be comparable in both sensitivity and repro- ducibility. The SFE method was also rapid and almost solvent free.

3.2.3 Steam distillation

  Most of the compounds that cause taint or off-flavour can be assigned to the class of volatile flavour components. The only difference is that the taste of the compound is unpleasant in the particular food in which it is found. Some compounds, such as 2,4,6-trichloroanisole, which gives a ‘musty’ taint, are unpleasant in any food. Other compounds, such as guaiacol, are regarded as having a positive impact in smoked food or in some whiskies but are seen as taints in vanilla ice cream or apple juice. Steam distillation has been used for the extraction of flavour and odour compounds for some considerable time. It relies on the volatility of those compounds, which co- distil with the steam, separating them from the non-volatile components,

  Instrumental methods in detecting taints and off-flavours 35

  which remain in the sample. The distillate can be treated in various ways. For example, in the analysis of guaiacol in wine (Simpson et al., 1986), the distillate was collected in a basic solution to retain the analyte. It was then acidified and extracted with dichloromethane, which was concentrated to a low volume for GCMS analysis. In an investigation into a taint in melons (Sanchez Saez et al., 1991) the distillate was collected in dichloromethane in an ice bath. The dichloromethane was then used to extract the tainting material from the collected water. Analysis identified the compound responsible as 4-bromo-2-chlorophenol.

  In order to avoid overheating the sample, which might lead to the break- down of thermally unstable compounds, distillation can be carried out at lower temperature by placing the sample under vacuum. In an investiga- tion of a phenolic taint in cheese, the sample was shredded and the curd removed (Mills et al., 1997). The remaining water and fat mixture was steam distilled, under vacuum, at a temperature of 60 °C. The distillate was collected in a cold trap and extracted with diethyl ether. The extract was analysed by combined HRGC–olfactometry (GCO) and GCMS. 2-Bromo- 4-methylphenol was identified as the probable cause of the taint. A similar method has been used to study geosmin in the Nile tilapia fish (Yamprayoon and Noohorm, 2000a; 2000b). Vacuum distillation can also been used to prepare large quantities of extract for analysis, for example, in the case of Mexican coffee which was found to have a mouldy or earthy taint (Cantergiani et al., 2001). A 100 g sample was vacuum distilled and around 200–300 ml of distillate was collected. This process was repeated five times and the distillates were pooled and extracted with dichloromethane, which was then concentrated to 0.5 ml for analysis. The extract was examined by GCO and fractions isolated by preparative HRGC. The fractions were analysed by GCMS and a number of compounds were identified, including geosmin, methylisoborneol and 2,4,6-trichloroanisole.

  A recent development of the technique has been the use of microwave- assisted steam distillation. In 1996 Conte and co-workers described a system for the extraction of geosmin and methylisoborneol from catfish (Conte et al., 1996a). Later that year the authors described improvements to the apparatus (Conte et al., 1996b).The sample was placed in a microwave oven, in a vessel linked to a C18 silica solid phase adsorbent cartridge. Internal standards of cis-decahydro-1-naphthol and endo-norborneol acetate were added and a stream of argon was passed over the tissue and through the cartridge while the microwave was operated at 40 power for 10 min. The volatile organic compounds were then eluted from the trap with ethyl acetate and analysed by GCMS, with detection limits in the sub-parts-per- billion range (Conte et al., 1996b). The use of the solid phase trapping reduced the solvent required in the analysis to 2 ml and the use of a microwave oven cut the extraction time to 10 min. The method was further refined by the addition of SPME extraction (Zhu et al., 1999). The distillate from the sample was collected in a vial and the target compounds were

  36 Taints and off-flavours in food extracted by SPME and injected into a GCMS for analysis. This completely

  eliminated the need for solvent and simplified the final stage of the extrac- tion, making it more rapid to use. The authors investigated the recoveries of analytes using both methods.They concluded that the original solid phase trapping produced better recoveries, possibly because it was a closed system where all of the material distilling from the sample passed through the trap. In the other method, collection of the distillate took place in an open system, with the potential for evaporative loss of some of the distillate and the target compounds. The authors concluded, however, that use of SPME did not result in any loss of sensitivity or reproducibility.

  A similar apparatus was used for the analysis of geosmin and methylisoborneol in catfish (Lloyd and Grimm, 1999). Deuterated analogues of the target compounds were added to fish tissue and the volatiles were distilled out using microwave-assisted distillation with nitrogen as a purge gas. Again SPME was used to extract the analytes from the distillate for GCMS analysis. The authors studied the effects of a number of para- meters, including the operating power of the oven and the flow of the purge case, in order to optimise the extraction. They also compared the procedure with microwave distillation–solvent extraction and purge and trap–solvent extraction. They concluded that the SPME extract of the distillate produced the largest response from the GCMS analysis and that it gave good repro- ducibility. They reported a disadvantage of SPME in that it was only prac- tical to carry out a single GCMS analysis from the distillate. With solvent extraction, repeat injections were possible. Quantitative analysis of geosmin and methylisoborneol at the low part per trillion range was possible using this method (Grimm et al., 2000a). Grimm and co-workers also carried out analysis of catfish using a version of the apparatus in which no purge gas was used (Grimm et al., 2000b).

3.2.4 Combined steam distillation and solvent extraction

  One of the most popular methods for the extraction of volatile tainting com- pounds is SDE, usually using a variation of the apparatus first described by Likens and Nickerson (Likens and Nickerson, 1964). A review (Chaintreau, 2001) describes the original apparatus and some of the changes and improve- ments that have been made to it. The principle of operation of the Likens–Nickerson apparatus is straightforward.Water containing the sample is boiled in one flask while the extracting solvent is boiled in another. The vapours are allowed to mix and condense in a central chamber. The immis- cible condensates form two layers at the bottom of the central chamber and return pipes lead the water and solvent back to their respective flasks. Volatile compounds distil out of the sample with the steam and are extracted into the solvent during the mixing and condensing process. As the solvent passes back into its flask it carries the extracted volatiles with it. Over a period of time the volatile compounds will be extracted into the solvent.

  Instrumental methods in detecting taints and off-flavours 37

  Because the volatilisation of the solvent followed by the condensation, extraction and return form a cyclic process, a minimal amount of solvent can

  be used. It is possible to use relatively large sample sizes, since it is only the steam and the entrained volatiles that enter the body of the extractor. In the author’s laboratory, 100 g samples are regularly dispersed in 1 l of water and extracted using 30 ml of diethyl ether. After extraction the solvent can easily

  be concentrated to a low volume for analysis. Good recoveries can be obtained with this equipment. Whitfield reported recoveries of 80–100 for 2,4,6-trichloroanisole, 2,3,4,6- tetrachloroanisole and pentachloroanisole (Whitfield et al., 1986) and 93 recovery of 2,4,6-tribromoanisole (Whitfield et al., 1997) from both fruit and fibreboard. Recoveries do vary with the nature of the analyte. Steam dis- tillation and extraction of coffee suspensions showed similar recoveries of the anisoles but lower values for chlorophenols. The figures were 70–76 for the 2,4,6-trichlorophenol, 35 for 2,3,4,6-tetrachlorophenol and 18 for pentachlorophenol (Spadone et al., 1990). It is possible to improve the recoveries of basic or acidic compounds by changing the pH of the aqueous mixture. For example, in the quantitative analysis of 2,6-dibromophenol in crustacea the mixture of water and sample was acidified to pH 2 with sulfuric acid (Whitfield et al., 1988) to force the compound into its free acid form and increase its volatility. Recoveries can be adversely affected by a high lipid content in the sample (Au-Yeung and MacLeod, 1980).To counter this problem, a modification of the Likens–Nickerson apparatus used a steam generator to inject steam into the sample during extraction. This improved the recoveries of each of the compounds studied (Au-Yeung and MacLeod, 1980).

  A potential problem with SDE is the formation of artefacts during the extraction procedure. A comparison of SDE and vacuum distillation of hen meat (Siegmund et al., 1997) showed that thialdine could be detected in the extract from SDE but not in the extract from vacuum distillation. Investi- gation of the process showed that thialdine was being produced during the SDE extraction. One approach to reducing artefact formation has been the use of a Likens–Nickerson extractor modified to operate at reduced pres- sure (Chaintreau, 2001). Vacuum SDE was used to extract geosmin from beans, with the aqueous mixture boiling at 45°C (Buttery et al., 1976). Another version of a vacuum Likens–Nickerson apparatus was also described by Schultz and co-workers (Schultz et al., 1977). Operation under vacuum requires the use of a relatively non-volatile extracting solvent to avoid losses during the extraction (Chaintreau, 2001).

3.3 Headspace extraction

  Solvent-based extractions are not suitable for the analysis of very volatile compounds. The compounds can be lost during the extraction and clean-up

  38 Taints and off-flavours in food stages and those that are present in the final extract can be impossible to

  detect as they are masked by the presence of the extracting solvent. An alternative approach to analysis of these compounds is to use their volatil- ity as part of the separation step, collecting them from the headspace of the sample. Headspace techniques have the advantage that they can be carried out at relatively low temperatures, minimising the possibility of artefact formation and they require little or no solvent, depending on the method in use.

3.3.1 Static headspace sampling

  Static headspace sampling is the simplest of the available techniques. A portion of the sample is placed in a sealed container and allowed to stand until the concentration of volatile compounds in the headspace above the sample has reached equilibrium. The sample can be heated to increase the concentration of analytes in the headspace. A volume of the headspace is then removed and introduced into the analytical instrument. The sensitiv- ity of this approach is governed by the volume of headspace taken from the sample and the concentration of the target compounds in the headspace at the time of sampling.

  Sorbic acid, which is used as a preservative, can be decarboxylated by some species of Penicillium mould to produce 1,3-pentadiene which has an odour described as ‘paint’, ‘kerosene’ or ‘plastic’ (Marth et al., 1966). Static headspace has been used for the analysis of 1,3-pentadiene in cheese spread (Daley et al., 1986), margarine and cheese (Sensidoni et al., 1994). It was also used, in conjunction with GCMS and combined gas chromatogra- phy–infrared spectroscopy (GCIR), in the investigation of a musty taint in packaging film (McGorin et al., 1987). The simplicity of static headspace sampling lends itself to automation and several commercial headspace injectors are available for HRGC instruments. In an investigation of autox- idation in wheat germ, hexanal was chosen as a marker compound for the development of off-flavour and was measured in a number of samples using

  a headspace autosampler.

3.3.2 Dynamic headspace sampling

  In order to overcome some of the sensitivity limitations of static headspace sampling, a dynamic headspace technique, also known as purge and trap sampling, has been used. The sample is placed in a container fitted with a purge head connected to a trap. A clean inert gas is passed over a solid sample or through a liquid sample and allowed to exit through the trap. Volatile compounds are swept out of the headspace into the trap. Since the concentration of the analyte in the headspace is being depleted, the equi- librium is disturbed and more of the material will be transferred from the sample to the headspace and so, as the sampling is continued, the amount

  Instrumental methods in detecting taints and off-flavours 39

  of the analyte held on the trap will increase. As with static headspace, heating the sample can increase the concentration of the compounds in the headspace and so increase the amount of sample on the trap.

  For any particular compound the limiting factor of the trapping process is the breakthrough volume. As the purge gas passes through the trap it will act in the same way as the carrier gas in a gas chromatographic column. At some point the analyte will reach the end of the trap and begin to be eluted. From then onwards there will be a loss of material from the trap. The break- through volume will depend on the analyte, the material used in the trap, the purge gas and the temperature. The use of several different trapping materials has been reported in the literature. Horwood and co-workers used Poropak Q for the determination of 1,3-pentadiene in cheese (Horwood et al., 1981), while Chromosorb 105 was used for the analysis of trimethy- larsine in prawns (Whitfield et al., 1983). A widely used trapping medium is Tenax, which is available in several forms. Tenax GC was used to analyse for the ethyl esters which were causing a ‘fruity’ off-flavour in milk (Wellnitz-Ruen et al., 1982), for styrene in apple juice (Durst and Laperle, 1990) and in the investigation of a musty taint in French fries (Mazza and Pietzak, 1990). Tenax TA was used for the identification of geosmin in clams (Hsieh et al., 1988) and in cultured microorganisms (Börjesson et al., 1993) and in the investigation of off-flavours generated in olive oil during oxida- tion (Morales et al., 1997). Tenax TA was compared with three carbon-based trapping materials for the analysis of off-flavours generated in whey protein concentrates during storage (Laye et al., 1995). It was concluded that Tenax TA was the best material for some of the compounds under study, while Vocarb 3000 was best for the rest of the compounds.

  Generally, after the sampling phase is complete, the trapped volatiles are introduced into a HRGC column using some form of thermal desorption and cold-trapping system. The trap, containing the volatiles, is placed in an oven connected to a cold trap, which is connected in turn to the head of the HRGC column. Carrier gas passes through the adsorbent, which is heated to release the volatiles, which are then concentrated in the cold trap. After

  a suitable purge time, the cold trap is rapidly heated to transfer the volatiles to the HRGC column in a narrow band and the analysis is started. If the original sample contained water there is a possibility that moisture would

  be retained in the trap during the sampling process. This can lead to prac- tical difficulties during the thermal desorption and cold-trapping injection. The water can be eliminated by purging the trap with clean dry gas after the sampling period is complete (Ramstad and Walker, 1992). The break- through volume for water in most trapping materials is considerably lower than those for the volatiles of interest and so water can be removed while retaining the analytes. Solvent extraction has also been used to remove the volatiles from the trap (Harris et al., 1986; Mazza and Pietzak, 1990). This has the advantage that more than one analysis of the trapped compounds can be undertaken but the corresponding disadvantage that the sensitivity

  40 Taints and off-flavours in food of the analysis will be lower, since a fraction of the total sample is being

  analysed. Solvent extraction can only be used for the analysis of those com- pounds that are well separated from the solvent during analysis. Volatile compounds can be analysed directly from non-aqueous samples using thermal desorption and cold trapping without a preliminary sampling stage. In an analysis of geosmin in fish, oil extracted from the fish was placed onto

  a plug of glass wool in the injection port of a gas chromatograph, while the head of the column was cooled to -30°C. After the volatiles had condensed on the head of the column, the temperature was raised and the analysis started (Dupuy et al., 1986).

  A variation on dynamic headspace sampling is the closed-loop stripping apparatus (CLSA) first described by Grob and co-workers (Grob, 1973; Grob and Zücher, 1976). Here the sample and the trap are part of a loop, which also contains a pump. The gas in the loop, either air or an inert gas, is continuously pumped through the system. Volatiles are purged from the sample and concentrated in the trap, which contains a small amount of a carbon-based adsorbent. As with an open system the limiting step will be the breakthrough volume of the analytes. Since this is a cyclic system, com- pounds eluting from the end of the trap will not be lost but will be carried back through the system in the gas stream. Thus the concentration of the compounds on the trap will reach equilibrium. After sampling is complete, the volatiles can be removed from the trap in a very small amount of solvent. It is possible, therfore, to analyse the extract of each sample more than once, with the disadvantage that the sensitivity is correspondingly reduced compared to a single-shot technique such as Tenax trapping. The extraction efficiencies of different solvents for a wide range of volatiles have been investigated (Borén et al., 1985). Closed-loop stripping apparatus has been used in the investigation of musty taints in water (Krasner et al., 1983; Martin et al., 1988) and in beet sugar (Marsili et al., 1994). It was used for the analysis of geosmin and 2,4,6-trichloroanisole in a study on the effec- tiveness of biological treatment in removing odorous compounds from water (Huck et al., 1995). Objectionable odours from the Lake of Galilee were identified using CLSA and GCMS and found to be caused by organosulfur compounds such as dimethyl sulfide, dimethyl disulfide, dimethyl trisulfide and methane thiosulfonate (Ginzburg et al., 1998).

3.4 Solid phase microextraction (SPME)

  Since its introduction by Pauliszyn and co-workers (Arthur and Pauliszyn, 1990; Arthur et al., 1992; Potter and Pauliszyn, 1992) SPME has become widely used as an extraction technique. The principles of its operation are simple. A fused silica fibre, coated with one of a number of polymers, is introduced into the headspace above a sample or, for liquids, into the sample itself. Compounds from the material are adsorbed onto the fibre

  Instrumental methods in detecting taints and off-flavours 41

  until equilibrium is reached between the adsorbent and the matrix. The fibre is then removed from the sample and introduced into the injection port of a gas chromatograph where the trapped compounds are thermally desorbed into the head of the column for analysis. A review covers the theory of operation in some detail (Pauliszyn, 2000). The equilibrium between the sample and the fibre is influenced by a number of factors and so the choice of fibre and sampling conditions can affect the recovery of the analytes and therefore the sensitivity of the method. A second review (Shirey, 2000) describes the optimisation of these parameters.

  The SPME technique has a number of advantages. It is solvent-free, extraction is both simple and rapid and injection into the gas chromato- graph requires no specialised equipment. The operations involved in sam- pling and injection are very similar to those of a static headspace injection. As a result the technique has proved amenable to automation and com- mercial systems are available which will carry out unattended extraction and analysis. A comparison between SPME and dynamic headspace analy- sis using Tenax, concluded that with a polydimethylsiloxane (PDMS) fibre, the two methods recovered similar compounds with comparable repro- ducibilities but that the Tenax analysis was more sensitive (Elmore et al., 1997). On the other hand, in a study of light-induced lipid oxidation prod- ucts in milk, it was concluded that SPME was the better technique and that it was cheaper to operate in practice (Marsili, 1999). Three different types of fibre were compared in the development of a method to analyse hexanal and pentanal as markers for the development of off-flavours in cooked turkey. The phases used were CarboxenPDMS, PDMSdivinylbenzene (DVB) and DVBCarbowax. For these compounds, the PDMSDVB fibre showed the best combination of reproducibility, sensitivity and linearity (Brunton et al., 2000). Derivatisation of the compounds trapped on a fibre can be carried out before GCMS analysis. Diazomethane has been used to methylate chlorophenols and other acidic compounds by exposing the fibre, after sampling, to the reagent in a gaseous form (Lee et al., 1998).

  An area where SPME is widely used is in the analysis of 2,4,6- trichloroanisole in wines and corks. Evans and co-workers developed a SPME method for the compound in wine, which was comparable in per- formance to existing methods but was much more rapid (Evans et al., 1997). During the development of a method for the analysis of the compound in both wine and cork (Fischer and Fischer, 1997), the fibre was used to sample the headspace above either wine or ground cork. A comparison was made between sampling by immersion into the wine and sampling in the head- space. It was found that a better recovery of 2,4,6-trichloroanisole was obtained from the headspace. It was also suggested that dipping the fibre into the wine might lead to the transfer of non-volatile material to the injector of the gas chromatograph resulting in loss of performance. An automated analysis, using SPME in combination with GCMS, has also been described (Butzke et al., 1998). In all of these methods the detection An area where SPME is widely used is in the analysis of 2,4,6- trichloroanisole in wines and corks. Evans and co-workers developed a SPME method for the compound in wine, which was comparable in per- formance to existing methods but was much more rapid (Evans et al., 1997). During the development of a method for the analysis of the compound in both wine and cork (Fischer and Fischer, 1997), the fibre was used to sample the headspace above either wine or ground cork. A comparison was made between sampling by immersion into the wine and sampling in the head- space. It was found that a better recovery of 2,4,6-trichloroanisole was obtained from the headspace. It was also suggested that dipping the fibre into the wine might lead to the transfer of non-volatile material to the injector of the gas chromatograph resulting in loss of performance. An automated analysis, using SPME in combination with GCMS, has also been described (Butzke et al., 1998). In all of these methods the detection

  

  The SPME technique has been used for the analysis of taints in a wide variety of materials. An off-flavour in margarine was identified as a series of ketones produced by a mould which had grown in the product (Hocking et al., 1998). An off-flavour in milk, described as fruity and rancid, was shown to be due to a mixture of ethyl esters and fatty acids again formed by a microorganism which had contaminated the milk (Whitfield et al., 2000). The flavours and off-flavours formed during storage of strawberry juice were determined by headspace SPME (Golaszewski et al., 1998) and the compounds responsible for the off-flavours were identified. Investiga- tion of an off-flavour in a dressing showed the origin to be in one of the ingredients, beef plasma protein (Koga et al., 2001). The volatiles from the protein were analysed using SPME and GCMS. The compounds identified included hexanal, pentanal and 2-methylbutanal which are associated with oxidative rancidity. The SPME technique has been used to quantify acetaldehyde in spring water and various types of milk in a study to deter- mine the flavour threshold of the compound in these products (Van Aardt et al., 2001). Geosmin and 2-methylisoborneol have been determined in water using SPME and GCMS (McCallum et al., 1998; Lloyd et al., 1998). The technique has also been used in conjunction with microwave distilla- tion for the analysis of geosmin and 2-methylisoborneol in fish. The distil- late from the fish tissue was analysed using SPME rather than solvent extraction (Zhu et al., 1999; Grimm et al., 2000a; Grimm et al., 2000b).

3.5 Gas chromatography and other methods

3.5.1 Gas chromatography–olfactometry (GCO)

  Most investigations of taint or off-flavour rely on gas chromatography. There are numerous reviews and books discussing the fundamentals of the technique and its application to food chemistry (Siouffi and Delaurent, 1996; Le Bizec, 2000; Handley and Alard, 2001). When the tainting com- pounds are known, it is possible to use a selective detector, such as ECD or a mass spectrometer operating in SIM mode, to increase the specificity and sensitivity of the analysis. When the causes of the taint are not known they may be identified using GCMS, with the mass spectrometer operating in scan mode to collect spectra of the compounds eluting from the column. Sometimes it is possible to compare the total ion chromatogram (TIC) from the GCMS analysis of a tainted sample with that of an untainted sample and observe differences which correspond to the tainting compounds. More frequently, however, the responses for the taints are small and the differ- ences are not obvious. In this case the investigation can be easier if the retention times of the tainting compounds can be found, allowing the analyst to concentrate on those areas of the TIC. One popular approach

  42 Taints and off-flavours in food

  Instrumental methods in detecting taints and off-flavours 43

  to identifying the retention characteristics is to use GCO, where the human nose is used as a selective and sensitive detector of the compounds of interest.

  This technique has been widely used for some time in the analysis of flavour compounds, including tainting compounds, and the principles of the technique have been extensively reviewed (Acree, 1993; Blank, 1997; Feng and Acree, 1999; Friedrich and Acree, 2000). The basic idea of GCO is simple. The effluent from the gas chromatographic column is mixed with air and water vapour and led to a point where a human assessor can perceive the aromas of compounds eluting from the column. The entire effluent of the column can be used or a splitter can be inserted, to divert part of the effluent to the olfactometer and the rest to a detector, for example a mass spectrometer. There are a variety of parameters to be considered in the optimisation of the technique. For example, the flow of make-up air and the humidity need to be controlled to present the compounds to the observer in the most efficient manner (Hanaoko et al., 2000). The most significant factors, however, are those that affect the perception of aroma by the sensory panellist. These include the alertness of the panellist, the tiredness of the senses, olfactory adaptation and the ability of the assessor to detect particular compounds (Kleykers and Scifferstein, 1995; Friedrich and Acree, 2000;Van Ruth and O’Connor, 2001). Given the short elution time of a com- pound from the HRGC column, even the rate of breathing of the subject can affect the detection of the odour (Hanaoka et al., 2001).

  Gas chromatography–olfactometry has also been used in the investiga- tion of taints. For example, it was used by Patterson in 1970, during the investigations into taints in meat (Patterson, 1970). A selection of reports describes the use of GCO for the analysis of 2,4,6-trichloroanisole in coffee (Holscher et al., 1995), 2-bromo-4-methylphenol in cheese (Mills et al., 1997) and 2-aminoacetophenone in wine (Rapp, 1998). It has been used for the study of tainting compounds in cork (Moio et al., 1998) and for off-flavours in boiled potatoes (Petersen et al., 1999) and in soybean lecithins (Stephan and Steinhart, 1999). Darriet and co-workers described the use of GCO to identify geosmin as the cause of an earthy taint in wine (Darriet et al., 2000), while Escudero et al. (2000) used GCO to identify methional as the source of a vegetable off-flavour in wine.

3.5.2 Gas chromatography with selective detectors

  In cases of taint or off-flavour when the identities of the tainting compounds are known and the molecules contain a suitable heteroatom, it is possible to carry out a gas chromatographic analysis using a selective detector. The detector can be chosen so that it will produce a response for the target compound but not for compounds which do not contain the heteroatom. This reduces the potential interference from any co-extracted compounds that elute at the same retention time as the tainting compounds. When the

  44 Taints and off-flavours in food compound contains a halogen such as chlorine or bromine, it is likely to

  produce a signal on an ECD. The greater the number of halogens in the molecule the greater the response produced. An ECD has been used in the analysis of chlorophenols and chloroanisoles. In a study on Mason jars, pen- tachlorophenol was found to be present in the jar lids and to have migrated into various foods stored in the jars using home canning techniques (Heikes and Griffitt, 1980). The pentachlorophenol was extracted from the samples using dichloromethane at acid pH and methylated to form the penta- chloroanisole, which was then determined by HRGC–ECD. Chlorophenols and chloroanisoles in wines and corks have been analysed by solvent extraction followed by HRGC–ECD (Bayonove and Leroy, 1994), as have chloroanisoles and chloroveratroles in water (Paasivirta and Koistinen, 1994), and chloroanisoles in pharmaceutical products (Ramstad and Walker, 1992).

  Sulfur-containing compounds are present in many different foods and contribute to their flavour. For example, dimethyl sulfide contributes to the taste and flavour of beer (Scarlata and Ebeler, 1999). In some cases, however, an excess of one or more of the sulfur compounds can lead to an off-flavour. Analysis of sulfur compounds in various foodstuffs has been carried out using three types of element selective detector, the flame photometric detector (FPD), the sulfur chemiluminescence detector and the atomic emission detector. Mistry and co-workers compared these detec- tors and concluded that the atomic emission detector showed a greater sen- sitivity and a wider dynamic range than the other two (Mistry et al., 1994). The atomic emission detector has been used to determine dimethyl sulfide and its precursors in red wine (Swan, 2000) and also for the analysis of chlorophenols, at parts per trillion levels, in drinking water (Turnes et al., 1994).

  Analysis of sulfur compounds produced during UHT (ultra high tem- perature) treatment of milk (Steely, 1994) and of sulfur compounds in brandy (Nedjma and Maujean, 1995) has been carried out by HRGC com- bined with the chemiluminescence detector. A study of its use in wine examined the optimisation of the operating parameters and concluded that it was more sensitive for volatile sulfur compounds than the FPD (Lavigne- Delcroix et al., 1996). A review on the analysis of sulfur compounds in wine describes the use of both chemiluminescence and the FPD (Mestres et al., 2000). The FPD detector has been used by Mestres and co-workers to analyse volatile sulfur compounds in wine using static headspace extrac- tion (Mestres et al., 1997) and headspace SPME (Mestres et al., 1998; 1999a; 1999b). They concluded that a commercially available fibre, CarboxenPDMS, was the most suitable for this analysis (Mestres et al., 1999b). The combination of headspace SPME and chromatography with an FPD has also been used to analyse beer for dimethyl sulfide (Scarlata and Ebeler, 1999) and a number of other volatile sulfur compounds (Hill and Smith, 2000). The FPD has also been used in the analysis of dimethyl

  Instrumental methods in detecting taints and off-flavours 45

  sulfide in rice and its products (Ren et al., 2001) as well as in heat-treated milk (Bosset et al., 1996).

3.5.3 Gas chromatography–mass spectrometry

  The combination of gas chromatography and mass spectrometry has resulted in the most powerful tool available for the analysis of volatile organic compounds. There are many books available both on the subject of mass spectrometry and GCMS. Examples include those by De Hoffmann and Stroobant (2001) and by Hubschmann (2001).

  When a MS is used in conjunction with a gas chromatograph there are two common modes of operation, full scan and SIM. In full scan analysis the mass spectrometer is set to scan continually over a given mass range and record complete mass spectra. If the total signal in each spectrum is summed and these summed values are plotted against time, the result is a TIC, which resembles the output from a non-specific GC detector such as an FID. The mass spectrum at any point in the trace can be examined and searched against existing libraries of spectra that are usually included with the data system of any GCMS instrument. The spectrum of an unknown can also provide information on the structure of that compound. Full scan mass spectrometry is often used to carry out a survey analysis, identifying the volatile compounds present in a sample. In situations where enough material is present in the extract, full scan GCMS can also give a very selec- tive quantitative assay for a target compound, since the combination of mass spectrum and retention time gives a high degree of confidence in the identification.

  Selected ion monitoring mass spectrometry is used for the identification and quantitation of known compounds. The MS is set up to record data from a small number of ions, characteristic of the target compounds, rather than scanning the entire mass range. The result is an improvement in sen- sitivity of the analysis, with a minimal loss in the selectivity. In SIM analy- sis, the identification of a compound is based on the elution of the target ions at the correct retention time. Additionally the ratios of the peak areas due to each ion should be the same in the target compound as in a refer- ence standard. A good example of the use of both full scan and SIM was in the work carried out on 2,6-dibromophenol in prawns (Whitfield et al., 1988). The initial analyses used full scan GCMS to identify the tainting compound. Subsequent quantitative analyses used SIM, allowing a much simpler sample preparation to be used, because of the improved sensitivity of the analytical method.

  GCMS analysis allows the use of compounds containing stable isotopes as internal standards. For example, some or all of the hydrogen atoms in

  the molecule can be replaced with deuterium; alternatively, 13

  C can be used

  to replace some or all of the 12

  C atoms. The result is a stable molecule that

  has chemical and physical properties very similar to that of the original

  46 Taints and off-flavours in food compound. The labelled molecule is very unlikely to occur naturally, so

  there is little risk of interference from molecules of the internal standard naturally present in a sample. Additionally, the mass spectrum of the stable isotope shows the same breakdown pattern as that of the original molecule but the fragments have different masses. The shift in the mass will depend on the number of altered atoms in the fragment. In a SIM analysis the same fragments can be monitored in the target compound and the standard by setting up two masses for each fragment. The disadvantage to the use of any internal standard is that it does reduce the time that the mass spec- trometer spends detecting each ion, since more ions are being detected in the same time; this can lead to a drop in sensitivity.

  Deuterated toluene has been used as an internal standard for the analysis of a range of volatile compounds by purge and trap GCMS (Zhou et al., 2000), making use of the fact that the compound is unlikely to occur naturally. Stable isotopes have been used in the analysis of 2- aminoacetophenone in wine (Dollman et al., 1996), sulfur-containing com- pounds in wine (Kotseridis et al., 2000) and 2,4,6-trichloroanisole in cork stoppers (Taylor et al., 2000).

  The analysis of 2,4,6-trichloroanisole in wine has also been carried out using an isomer of the compound as an internal standard (Aung et al., 1996). Here, 2,4,5-trichloranisole was used. This compound is not known to occur naturally, so there is little risk of interference from the samples. The inter- nal standard has the same fragment masses as the target compound but a different chromatographic retention time. Both compounds can be detected in SIM, using the same target ions, without any reduction in the sensitivity of the analysis.

  Some mass spectrometers can be used to distinguish between masses that differ by small amounts, which allows them to make use of the differences in masses that are due to the elements present in a fragment. The masses of the atoms of each element are not whole numbers, except for that of the most prevalent isotope of carbon, which has been set to 12 and from which all other elements are measured. For example, the mass of hydrogen is 1.00794, so compounds containing only carbon and hydrogen have masses that are greater than integers, i.e. mass sufficient. Some elements, such as the halogens, have masses that fall just below the nearest integer, so a compound containing enough of these elements will have a mass just below the integer value, i.e. it will be mass deficient. For example, tetradecene has a mass of 196.21910, while trichlorophenol has a mass of 195.92495.

  These properties can be used to detect halogenated compounds in a sample. If full scan GCMS is carried out with a mass spectrometer capable of distinguishing between the mass deficient and mass sufficient fragments, the data from the analysis can be inspected to determine if any mass defi- cient species are present. The data systems produced by some instrument manufacturers allow this to be done automatically by applying filters to the

  Instrumental methods in detecting taints and off-flavours 47

  data from the analysis. For example, a total ion chromatogram or a mass spectrum could be generated using only ions with fractional masses between 0.7 and 0.99. This would screen out all the mass sufficient ions in the analysis, leaving data only from potentially halogenated compounds. These techniques have been used in the author’s laboratory to identify brominated and iodinated phenols and cresols.

  An alternative approach to the identification of halogenated species is the use of negative ion chemical ionisation (NCI). Under these conditions many halogenated species produce a fragment consisting of a negatively charged halogen atom. It is possible to use reconstructed ion chro- matograms of the halogen ions to examine scan data or to carry out the analysis using the ions in SIM mode (Sanders and Sevenants, 1994). The combination of these techniques has also been successfully used in the author’s laboratory to identify brominated and iodinated compounds. Chemical ionisation can also be carried out in the positive mode (PCI).

  A combination of electron impact and positive chemical ionisation has been used in the analysis of geosmin and 2-methylisoborneol in water (McCallum et al., 1998).

3.5.4 High-performance liquid chromatography (HPLC)

  Although most tainting compounds are amenable to HRGC-based methods, there are some cases where HPLC is a more suitable method. For example, in the analysis of indole and skatole in pigs, HPLC has been used as the basis of a simple and rapid assay (Tuomola et al., 1996; Garcia- Regueiro and Ruis, 1998; Ruis and Garcia-Regueiro, 2001). It has also been used to study the formation of 4-vinylguaiacol in orange juice (Lee and Nagy, 1990).

  In cases where the tainting compounds are large or polar molecules, HPLC is the only method available for their analysis. Off-flavours in milk, produced by the Maillard reaction between lactose and lysines, have been studied using HPLC with a diode array detector and HPLC coupled to a mass spectrometer (LC-MS). Tentative identifications of some of the prod- ucts were possible (Monti et al., 1999). Development of off-flavours in fatty foods can be the result of oxidation of the lipids to volatile compounds such as aldehydes and ketones. This reaction proceeds by the formation of hydroperoxides which then break down to produce the smaller compounds. The formation of the hydroperoxides has been studied using methods based on HPLC (Akasaka et al., 1999).

3.6 Stir-bar sorptive extraction

  Stir-bar sorptive extraction (SBSE) is a technique based on the principles of SPME. The sampler is a small stirrer with a bar magnet encased in glass.

  48 Taints and off-flavours in food The surface of the glass is coated with a polymer in a similar way to the

  surface of a SPME fibre. As the bar is used to stir an aqueous sample, organic compounds partition between the sample and the polymer. After sampling is complete the bar is removed, washed with clean water and care- fully dried. The bar is then inserted into a thermal desorption and cold trap- ping injector which transfers the trapped compounds onto the head of a GC column for analysis. The system has all the advantages of SPME but with a greater capacity for organic compounds, owing to the increase in the surface area of the adsorbent. It has been used to analyse geosmin, methylisoborneol and 2,4,6-trichloroanisole in drinking water (Ochiai et al., 2001) with detection limits below 1 ppt. It has also been investigated as a tool for the determination of various types of taint in wine, packaging and breakfast cereal (Offen et al., 2001).

3.7 Electronic noses

  Electronic noses are a relatively recent development, with commercial applications of the technique starting to appear in the mid-1990s. A good introduction to the field is given in the book by Gardner and Bartlett (1999). Learning from the functional principles of the human nose, the elec- tronic nose comprises an array of electronic chemical sensors with broadly overlapping specificities and an appropriate pattern recognition system capable of recognising simple or complex odours (Persaud and Dodd, 1982; Persaud and Travers, 1997).

3.7.1 Principles of electronic noses

  The principle of operation of electronic noses is simple. Headspace gas from

  a sample is introduced to an array of sensors, which will respond to differ- ent degrees depending on the nature of the organic compounds present. The result is a histogram with each sensor represented by a vertical bar whose height is the intensity of response. In some ways, this resembles a mass spectrum and, in fact, mass spectrometers are the basis of some elec- tronic noses. The data from the histogram of responses from different samples are then assessed statistically and differences between the samples can be plotted. Electronic noses can also be trained to detect certain types of odour pattern.

  Current systems generally employ fewer than 50 sensors and are still far removed from a true ‘electronic nose’ with the full complexity of the human

  7 nose with its 10 8 to 10 receptors and part of the central nervous system for signal processing. However, the application-specific electronic nose

  (ASEN) with sensors and algorithms specialised for a specific application offers a cost-effective option for the on-line monitoring of flavours and off-flavours.

  Instrumental methods in detecting taints and off-flavours 49

3.7.2 Components of an application-specific electronic nose system

  Figure 3.1 shows a schematic diagram of the application-specific electronic nose (ASEN) system of odour detection. Components 1 to 3 of the system comprise the odour delivery system which provides control over the sample’s presentation to the sensor array. The array of sensors (component

  4) comprises sensors of one or more transducer types as base devices. These carry the chemically sensitive layers selected for the application.

  For aroma assessment, the sensors most frequently employed are poly- meric chemoresistors. For applications such as the identification of

  Sample handling

  Filter or

  (may include

  enzyme, …)

  modulation of

  (optional)

  concentration)

  4 5 Sensor electronics

  Array of

  with signal processing

  sensors with

  (averaging,

  integrated signal

  linearisation)

  conditioning

  Feature extraction (e.g. amplitude, slope,

  phase shift)

  fingerprint library

  vector X 1 recognition

  sample

  (e.g. cluster

  classification

  analysis)

  Artificial neural network (ANN)

  Fig. 3.1 Components of an application-specific electronic nose (ASEN) system.

  50 Taints and off-flavours in food off-odours, metal oxide (MeOx) semiconductors are employed more fre-

  quently. The traditional material here is tin oxide (SnO 2 ) which is the basis

  of many commercial sensors for industrial and domestic monitoring of combustible gases. A more recent material in commercial MeOx sensors is

  Ga 2 O 3 . Other semiconducting metal oxides have been examined for their gas sensing characteristics (Lampe et al., 1996; Kohl, 1997; Capone et al., 2000). The sensitivity characteristics of MeOx gas sensors are tuned by varying the preparation of the material (doping and sintering), the contact area (Hoefer et al., 1997) andor the operating temperature. They respond

  to gases and volatiles such as H 2 , CH 4 , NH 3 , CO, NO x , SO x ,H 2 S and alco-

  hols. Their molecular receptive range is more limited than that of polymeric chemoresistors. Less frequently used are the organic semiconductors such as the phthalocyanines (Huo et al., 2000). An interesting approach is the use of the intact chemoreceptors of insects as odour sensors. Some arthropods have extraordinary sensory abilities and these can be harnessed by inter- facing their chemoreceptive organs to microelectronic devices (Schütz et al., 1996; 1997). Blowfly receptors coupled to microelectrodes detected 1,4- diaminobutane in the range 1 ppb to 100 ppm (decaying meat odour), butanoic acid between 20 ppm and 200 ppm and 1-hexanol from 8 ppm to 500 ppm (Huotari, 2000).

  A combination of transducer types and sensing layer types (hybrid sensor systems) can be employed within an ASEN system. The molecular receptive range (MRR) can be widened and different aspects of the same odour molecule can be captured. As an example, functional groups can be probed with conducting polymers, molecular mass can be measured with piezoelectric devices, steric selectivity can be achieved with lipid layers or through functional side groups of polymers. In some odour-sensing appli- cations, it can also be meaningful to include a specific chemical sensor such as a biosensor with an immobilised enzyme, an immunosensor with an immobilised antibody or a DNA sensor.

  The result after processing by component 5 is an output in the form of the feature vector (component 6). This feature vector is fed into the evalu- ation module (component 7) where it can now be plotted in the chosen graphical representation such as bar chart (Fig. 3.2) or web chart (Fig. 3.3) and compared to a fingerprint library of odours. Additionally, feature vectors for a population of odour samples can be analysed using pattern recognition techniques (PARC). Each sampled odour can be described by its principal components and clusters of odour samples falling into a spec- ified range for one or more principal components can be identified (Fig. 3.4). Alternatively, an artificial neural network (ANN) can be trained to interpret the pattern (Gardner and Hines, 1997). Neural networks can be combined with fuzzy logic (Berrie, 1997) into neuro-fuzzy systems (Theisen et al., 1998). Either way, the ASEN arrives at component 8, odour identifi- cation and classification of the sample in terms of flavour or off-flavour attributes.

  Instrumental methods in detecting taints and off-flavours 51

  Sensor number

  Fig. 3.2 Bar chart representation of the feature vectors for two aromas, (a)

  citronellol and (b) cineole (after Persaud and Travers, 1997).

  A new addition to the spectrum of electronic nose technologies is the fingerprint mass spectra system (FMS). This instrument (Dittmann et al., 1988; Shiers et al., 1999; Dittmann and Nitz, 2000) is based on a quadrupole mass spectrometer combined with a headspace sampler and a computer. Volatile sample components are introduced into the MS without separation thus creating a mass spectrometric pattern. This reduced mass spectromet- ric pattern is analysed with the pattern recognition techniques developed

  Cineole 6 Citronellol

  Fig. 3.3 Web chart representation of the vectors for the aromas citronellol and

  Citronellol Cineole

  –1.5

  Fig. 3.4 Patterns generated by cluster analysis of the aromas citronellol and cineole (the first principal component is plotted along the horizontal axis, the

  second along the vertical axis).

  Instrumental methods in detecting taints and off-flavours 53

  for the electronic noses as described previously. A database of known reference samples is created and this can be used later to identify the property of interest in a new sample. The use of the label ‘electronic nose’ for the FMS systems has apparently prompted some manufacturers of electronic noses (as in the original definition) to rename their system as sensor arrays technology (SAT) (Mielle and Marquis, 2000; Mielle et al., 2000). Mielle and co-workers note that the competition between the manu- facturers of these two types of systems – SAT and FMS – has been very aggressive since 1998. The respective advantages have been hotly debated. Supporters of the FMS systems emphasise their selectivity, adaptability and sensitivity. They concede that a well-equipped chemical analysis laboratory and some investment in time and effort would be required for the screen- ing stage prior to training and final measurement. Supporters of the SAT systems argue that the sensitivity and accuracy are comparable and empha- sise the smaller size and cost with the potential for further reductions in both size and cost. The issue of assay speed is a complex one. The sample throughput is related to variables such as sampling time, headspace equili- bration time, recovery or cleaning time.

3.7.3 Uses of electronic noses

  Evans et al. (2000) tested an electronic nose based on a conducting polymer array plus a neural network against a human sensory panel for the quality assessment of wheat with respect to the absence of taints. After training the system with 92 samples, a predictive success of 93 was achieved.

  Electronic noses have been used as a rapid screening method for taints in packaging (Pitt, 1996; Shiers and Squibb, 1999; Forsgen et al., 1999; Culter, 1999), for monitoring flavours and off-flavours in beer (Gardner et al., 1994; Pearce, 1994; Heberle et al., 2000) and for investigations into boar taint (Bourrounet et al., 1995; Annor-Frempong et al., 1998). Some workers have investigated the application of an electronic nose to the problem of conta- mination of corks by 2,4,6-trichloroanisole and have suggested that the system could provide a simple quality control tool (Rocha et al., 1998). There is some indication that electronic noses are beginning to be used in quality control. In the United Kingdom, a number of dairy companies are using electronic noses to screen incoming milk for taint, especially for chlorophenols and chlorocresols (Kilcast, 2002). Galdikas et al. (2000) used an array of eight MeOx sensors combined with a neural network to monitor chicken meat during storage. An electronic nose based on eight metallo- porphyrines coated onto quartz microbalances (QMBs) was tested in the monitoring of ageing cod fillets and veal (Di Natale et al., 1997).

  An alternative approach to sample introduction was described by Marsili (1999a; 2000), who used SPME to collect volatiles from the headspace above milk. The volatiles were rapidly desorbed and introduced to a mass spectrometer. The spectra from the samples were processed using

  54 Taints and off-flavours in food multivariate statistics. This system could discriminate between various types

  of off-flavour in milk (Marsili, 1999a) and could be used to give accurate predictions of milk shelf-life (Marsili, 2000).

  Other applications of electronic nose technology in this field include: • classification of food spoilage bacteria (Pavlou et al., 2000; Magan et al.,

  2001) • fish freshness (Di Natale et al., 1997) • meat freshness (Eklov et al., 1998; Blixt and Borch, 1999) • detecting off-odours in sugar (Kaipainen et al., 1997) • taints in beer (Pearce et al., 1993; Tomlinson et al., 1995).